Offshore
Photo
DAIR.AI
// Scaling Generalist Agent Intelligence //
This new research introduces AgentSkiller, a fully automated framework for synthesizing multi-turn interaction data across realistic, semantically linked domains. The architecture uses a Directed Acyclic Graph to ensure determinism and recoverability throughout the generation pipeline.
The pipeline follows five phases.
* First, it builds domain ontologies and Person-Centric Entity Graphs to structure data around realistic scenarios.
* Second, it defines Service Blueprints for MCP servers.
* Third, it populates environments with constraint-satisfying databases and strict domain policies.
* Fourth, a cross-domain fusion mechanism links services to simulate complex multi-turn tasks.
* Finally, execution-based validation filters out any trajectories that fail runtime checks.
The Dual-Model Architecture decouples semantic reasoning from code implementation. A Textual LLM handles domain expansion and policy formulation, while a Coding LLM handles SQL generation and Python implementation.
With approximately 11K synthesized interaction trajectories, AgentSkiller-14B achieves 79.1% on tau2-bench, outperforming GPT-o3 at 68.4% and Claude Sonnet 4 at 56.8%. On ACEBench-Agent, it scores 78.0%, surpassing Gemini-2.5-pro by 14.6 points.
AgentSkiller-4B at just 4 billion parameters scores 66.0% on tau2-bench, outperforming xLAM-2-70B at 41.0% and Qwen3-Thinking-235B at 58.6%. Cross-domain data is critical: adding it to single-domain training boosts tau-bench from 45.4% to 67.2%.
High-quality synthetic data with deterministic environment construction lets compact open-source models match or exceed proprietary systems on complex tool-use tasks.
Paper: https://t.co/6d92LkXBGm
Learn to build effective AI Agents in our academy: https://t.co/LRnpZN7L4c
tweet
// Scaling Generalist Agent Intelligence //
This new research introduces AgentSkiller, a fully automated framework for synthesizing multi-turn interaction data across realistic, semantically linked domains. The architecture uses a Directed Acyclic Graph to ensure determinism and recoverability throughout the generation pipeline.
The pipeline follows five phases.
* First, it builds domain ontologies and Person-Centric Entity Graphs to structure data around realistic scenarios.
* Second, it defines Service Blueprints for MCP servers.
* Third, it populates environments with constraint-satisfying databases and strict domain policies.
* Fourth, a cross-domain fusion mechanism links services to simulate complex multi-turn tasks.
* Finally, execution-based validation filters out any trajectories that fail runtime checks.
The Dual-Model Architecture decouples semantic reasoning from code implementation. A Textual LLM handles domain expansion and policy formulation, while a Coding LLM handles SQL generation and Python implementation.
With approximately 11K synthesized interaction trajectories, AgentSkiller-14B achieves 79.1% on tau2-bench, outperforming GPT-o3 at 68.4% and Claude Sonnet 4 at 56.8%. On ACEBench-Agent, it scores 78.0%, surpassing Gemini-2.5-pro by 14.6 points.
AgentSkiller-4B at just 4 billion parameters scores 66.0% on tau2-bench, outperforming xLAM-2-70B at 41.0% and Qwen3-Thinking-235B at 58.6%. Cross-domain data is critical: adding it to single-domain training boosts tau-bench from 45.4% to 67.2%.
High-quality synthetic data with deterministic environment construction lets compact open-source models match or exceed proprietary systems on complex tool-use tasks.
Paper: https://t.co/6d92LkXBGm
Learn to build effective AI Agents in our academy: https://t.co/LRnpZN7L4c
tweet
Offshore
Photo
NecoKronos
At this point, if you're still ignoring stacked imbalances after seeing my posts, you're actively choosing to make trading harder
It is literally the market showing you exactly where the unfinished business is.
If you have the correct template, these levels scream at you.
Stop ignoring the clearest alpha on your screen.
Wake up
#BTC
tweet
At this point, if you're still ignoring stacked imbalances after seeing my posts, you're actively choosing to make trading harder
It is literally the market showing you exactly where the unfinished business is.
If you have the correct template, these levels scream at you.
Stop ignoring the clearest alpha on your screen.
Wake up
#BTC
Stacked imbalances = The ultimate market magnets. 🧲
When aggressive orders stack up, they leave a trail of "unfinished business" that price is destined to return to.
In simple terms, an order flow imbalance happens when aggressive buyers or sellers completely overwhelm the other side at a specific price.
Stop guessing and start following the flow!
#BTC tweet
Offshore
Video
The Transcript
$SHOP:
tweet
$SHOP:
Shopify’s Q4 2025 earnings are out.
The headlines:
→ We now power more than 14% of the US ecommerce market
→ GMV is up 29% to $378B
→ Revenue is up 30% to $11.5B
→ Free cash flow exceeded $2B
→ Q4 was our first quarter above $3B in revenue
But 2025 wasn’t just about growth, it was about building for what’s next.
AI is changing how people shop. Merchants are racing to meet it. And we plan to keep powering them.
Watch the video for more. tweet
Offshore
Video
The Transcript
RT @tobi: it's going ok
tweet
RT @tobi: it's going ok
Shopify’s Q4 2025 earnings are out.
The headlines:
→ We now power more than 14% of the US ecommerce market
→ GMV is up 29% to $378B
→ Revenue is up 30% to $11.5B
→ Free cash flow exceeded $2B
→ Q4 was our first quarter above $3B in revenue
But 2025 wasn’t just about growth, it was about building for what’s next.
AI is changing how people shop. Merchants are racing to meet it. And we plan to keep powering them.
Watch the video for more. tweet
Offshore
Photo
Dimitry Nakhla | Babylon Capital®
RT @DimitryNakhla: Even off a very tough comp, where ratings revenue grew ~27% in Q4 2024, $SPGI still delivered ~12% growth in the latest Q4 2025 report. https://t.co/xXlrvpqTyJ
tweet
RT @DimitryNakhla: Even off a very tough comp, where ratings revenue grew ~27% in Q4 2024, $SPGI still delivered ~12% growth in the latest Q4 2025 report. https://t.co/xXlrvpqTyJ
tweet
Offshore
Photo
Clark Square Capital
RT @deusexdividend: These guys never #REF! out a model, don't need coffee runs or late night dinner orders, never push back on their VPs. It's never been more over https://t.co/uHgIq0tCHm
tweet
RT @deusexdividend: These guys never #REF! out a model, don't need coffee runs or late night dinner orders, never push back on their VPs. It's never been more over https://t.co/uHgIq0tCHm
tweet
Offshore
Video
Startup Archive
Jessica Livingston on investing in Stripe when the Collison Brothers were teenagers
Jessica reflects on a 19-year old Patrick Collison telling her and the other co-founders of Y Combinator that he wanted to take on the financial industry with his younger brother John.
“We were like, ‘Do you realize how hard this is? And you don’t have connections.’ But they were intrepid. They were like, ‘Well, we don’t have connections, but we’ll find connections.’”
She recalls their determination and focus:
“You think the head of a bank is going to take a 19-year old startup founder seriously? It seems pretty implausible, right? But they were good enough that they were able to convince these banks to work with them.”
Determination, Jessica argues, is “by far the most important quality. More than intelligence. More than previous success in school.” When she co-founded Y Combinator, the hypothesis was that they’d just fund all the best hackers from MIT and Harvard and they’d turn out to be great startup founders. But that turned out to not be true.
“Determination is the most important thing. Understanding your users and building a product with a great user experience is second most important.”
Jessica also believes being flexible minded is very important:
“You have this idea, you test it out, and it doesn’t always work the first time. You have to be able to say, ‘Okay, I thought I was going to do this, but let’s try this, even though I have a lot of energy vested in this, let’s try this direction. You really have to be open minded.”
And then you have to be convincing and a good leader:
“You are going to be convincing employees to join you. You’re going to be convincing investors to invest in you. When you get to the point where you’re doing deals with bigger companies, you have to convince them. Your whole world is convincing people, and so you have to be able to communicate your idea and convince people why they should care about you more than any of the other hundreds of startups out there.”
Video source: @ycombinator (2016)
tweet
Jessica Livingston on investing in Stripe when the Collison Brothers were teenagers
Jessica reflects on a 19-year old Patrick Collison telling her and the other co-founders of Y Combinator that he wanted to take on the financial industry with his younger brother John.
“We were like, ‘Do you realize how hard this is? And you don’t have connections.’ But they were intrepid. They were like, ‘Well, we don’t have connections, but we’ll find connections.’”
She recalls their determination and focus:
“You think the head of a bank is going to take a 19-year old startup founder seriously? It seems pretty implausible, right? But they were good enough that they were able to convince these banks to work with them.”
Determination, Jessica argues, is “by far the most important quality. More than intelligence. More than previous success in school.” When she co-founded Y Combinator, the hypothesis was that they’d just fund all the best hackers from MIT and Harvard and they’d turn out to be great startup founders. But that turned out to not be true.
“Determination is the most important thing. Understanding your users and building a product with a great user experience is second most important.”
Jessica also believes being flexible minded is very important:
“You have this idea, you test it out, and it doesn’t always work the first time. You have to be able to say, ‘Okay, I thought I was going to do this, but let’s try this, even though I have a lot of energy vested in this, let’s try this direction. You really have to be open minded.”
And then you have to be convincing and a good leader:
“You are going to be convincing employees to join you. You’re going to be convincing investors to invest in you. When you get to the point where you’re doing deals with bigger companies, you have to convince them. Your whole world is convincing people, and so you have to be able to communicate your idea and convince people why they should care about you more than any of the other hundreds of startups out there.”
Video source: @ycombinator (2016)
tweet
Offshore
Photo
Wasteland Capital
This is some candle. $SHOP
$30 intraday swing on a $110 stock / a $150bn market cap. Not exactly a micro cap. https://t.co/nwWQbmUoPr
tweet
This is some candle. $SHOP
$30 intraday swing on a $110 stock / a $150bn market cap. Not exactly a micro cap. https://t.co/nwWQbmUoPr
tweet
Offshore
Video
Quiver Quantitative
Lieu: Were there any underage girls at any party that Trump attended with Jeffrey Epstein?
Bondi: There is no evidence that Donald Trump committed a crime https://t.co/xHt2MVb03A
tweet
Lieu: Were there any underage girls at any party that Trump attended with Jeffrey Epstein?
Bondi: There is no evidence that Donald Trump committed a crime https://t.co/xHt2MVb03A
tweet
Offshore
Photo
Brady Long
RT @thisguyknowsai: 🚨 I watched a senior engineer at Anthropic build a feature in 4 hours that would've taken me 3 days.
He wasn't coding faster. He was running 8 Claude instances in parallel each solving different parts simultaneously.
The future of coding isn't writing code. It's orchestrating AI swarms.
Here's the framework:
tweet
RT @thisguyknowsai: 🚨 I watched a senior engineer at Anthropic build a feature in 4 hours that would've taken me 3 days.
He wasn't coding faster. He was running 8 Claude instances in parallel each solving different parts simultaneously.
The future of coding isn't writing code. It's orchestrating AI swarms.
Here's the framework:
tweet